Tech Tuesday: The A.I. Poker King and Watson's Ever Improving Tax Preparation Skills

This past year AI achievements have been stirring imaginations. So far the best chess player and best Go player have been beaten by "smart machines," and now the best poker players have fallen. What makes this story fascinating is that poker is a game that involves psychology, not simply logic. Can a machine really learn to bluff? Cade Metz's Inside Libratus, the Poker AI That Out-Bluffed the Best Humans* tells the story on Wired.com.

You might say we've been living in the "machine age" for quite some time, if you begin with the Eli Whitney's cotton gin. All throughout modern history machines have stirred both awe and insecurity. At the center of many of the fears is that the machines will take our jobs, or worse... replace the human race altogether. Nevertheless, as this story explains, we're fascinated by the strides they have made.

The computer in this instance had been developed by Carnegie Mellon. The high stakes game that 28-year Kim Dong, a world champion poker player, played in Pittsburgh was Texas Hold 'Em.

About halfway through the competition... Kim started to feel like Libratus could see his cards. “I’m not accusing it of cheating,” he said. “It was just that good.” So good, in fact, that it beat Kim and three more of the world’s top human players—a first for artificial intelligence.

To beat top Jeopardy players last year, Watson had to understand syntax and the words used in questions. For Google's DeepMind to win at Go it analyzed millions of players' moves before using this info to develop its skills by playing against itself. Libratus used a still different set of skills here. The article explains:

Libratus didn't need an ace up his sleeve.

Through an algorithm called counterfactual regret minimization, it began by playing at random, and eventually, after several months of training and trillions of hands of poker, it too reached a level where it could not just challenge the best humans but play in ways they couldn’t—playing a much wider range of bets and randomizing these bets, so that rivals have more trouble guessing what cards it holds. “We give the AI a description of the game. We don’t tell it how to play,” says Noam Brown, a CMU grad student who built the system alongside his professor, Tuomas Sandholm. “It develops a strategy completely independently from human play, and it can be very different from the way humans play the game.”

This was only part of the process it used. In addition to learning everything it could about poker before the match, Libratus also learned from the game itself. Perhaps a little like a good poker player who plays enough hands to learn his or her opponents' tells before moving into the big money plays.

What makes poker interesting, though, is the psychological aspect. Sometimes you raise a bet when you have no good cards in your hand. Sometimes you try to make the other player believe your bluffing so they stay in the game when you have four kings. Calculating odds is one thing, but getting inside the other players' heads is another.

What's next for Libratus? Wall Street? International trade negotiations.

Sunday evening a commercial in the Super Bowl showed us what Watson, the IBM A.I. will be up to this year. Watson's going to work on tax loopholes for us.

Meantime, life goes on...

* Read Inside Libratus, the Poker AI That Out-Bluffed the Best Humans here.